Structure and dynamics of molecular networks: a novel paradigm of drug discovery

6. Conclusions and perspectives 
 
The value of every drug design technology must be assessed by asking: “How 
much does the new technology help to solve one of the two central problems: the 
identification and validation of a disease-specific target or the identification of a 
molecule that can modify this target in a way that makes therapeutic sense?” (Drews, 
2003; Brown & Superti-Furga, 2003). We hope that we convinced the Reader that the 
network approach offers novel answers to both questions. In this concluding section 
we will highlight the major promises and perspectives of network-aided drug 
development.  
 
6.1. Promises and optimization of network-aided drug development 
One of the major promises of the network approach is its help to overcome the 
“one-effect/one-cause/one-target” magic bullet-type drug development paradigm 
(Ehrlich, 1908). Magic bullets do work – sometimes. When designing Strategy A-type 
drugs (Sections 4.1.1. and 4.1.7.), which target key nodes of the network to eliminate 
pathogens or malignant cells, an eradicating single hit may be beneficial. However, 
even here pathogen resistance or unexpected toxicity of anti-cancer drugs (and 
resistance against them) may ruin the final success. However, in the development of 
Strategy B-type drugs, which need to re-configure network dynamics from its disease-
affected state back to normal (Sections 4.1.1. and 4.1.7.), the traditional approach of 
rational drug discovery selecting a single and central target often fails. The paucity of 
disease-modifying anti-neurodegenerative drugs described in the preceding section is 
a sad example of the need for novel approaches in Strategy B-type drug design. 
We started our review with the statement that ‘business as usual’ is no longer an 
option in drug industry (Begley & Ellis, 2012). It is an especially warning message 
urging a radical change that the vast majority of new drugs are related to existing ones 
(Section 1.1.; Cokol et al., 2005; Yildirim et al., 2007; Iyer et al., 2011a). This 
situation justifies the saying of James Black that “the most fruitful basis for the 
discovery of a new drug is to start with an old drug” (Chong & Sullivan, 2007).  
Thus, a higher number of ‘surprisingly novel drugs’ is badly needed. How to find 
this ‘surprising novelty’? The failure of some efforts using the reductionist approach 
of rational drug design shifted the thinking to the other extreme saying that “we need 
unbiased research methods to cover complexity”. Indeed, unbiased machine learning 
methods successfully predict novel drug targets. However, artificial intelligence may 
miss true surprises (Section 2.2.2.). The network approach is often seen as another 
unbiased method. This leads to our first major conclusion, which we summarized 
in Fig. 19: the network approach must be combined both with human creativity 
and background knowledge.  
Network analysis does help in overcoming the ‘curse of spreadsheets’, and in 
comprehending the vast amounts of systems-level data, which became available in the 
last decade. However, network analysis does not make a miracle by itself. Originality, 
the highest level of human creativity (marked as the ‘surprise factor’ in Fig. 19), 
which strives for novelty, and identifies it in networks as the ‘prediction of the 
unpredictable’ (Section 2.2.2.) can not be missed. Similarly, we need comprehensive 
background knowledge to guide discovery (Valente, 2010). Combined with these two 
key assets, the current boom in network dynamics-related methods can help in 
discovering the truly surprising, novel actors of the cellular community, which are the 
hidden masterminds of cellular changes in health and in disease (Fig. 19). 

 
93
Our second major conclusion gives an optimized protocol of network-aided 
drug development suggesting alternating exploration and optimization phases of 
drug design (Fig. 19.). The discovery process has two major phases, the exploration 
phase and the optimization phase. During exploration a drive to discover the 
unexpected, playfulness and ambiguity tolerance are key assets (marked as the 
‘surprise factor’ on Fig. 19). In this exploration phase background knowledge may be 
temporarily suppressed. In contrast, in the optimization phase we need to suppress the 
playfulness and ambiguity tolerance of the exploration phase, and rank our previous 
options by the rigorous application of our background knowledge including all the 
well-orchestrated rules of the drug development process (Csermely, 2012; Gyurkó et 
al., 2012). Importantly, the sequence of exploration and optimization phases may be 
applied repeatedly, providing a more detailed ‘zoom-in’ of the optimal (drug) target 
than a single round of exploration/optimization (Fig. 19). The utility of repeated 
exploration/optimization rounds was shown by the thermal cycles of the well-known 
simulated annealing optimization method-induced ‘cooling’ and randomization-
achieved ‘heating’ (Möbius et al., 1997). 
The cyclic approximation of the discovery-optimum has consequences for the 
application of the network method itself. Recently, drug design-related networks have 
become increasingly complex. Based on the assumption that ‘everything is related to 
everything’ various types of datasets were increasingly mixed forming mega- and 
meta-mega-networks. More is not always better. Albert Einstein’s saying that “the 
supreme goal of all theory is to make the irreducible basic elements as simple and as 
few as possible without having to surrender the adequate representation of a single 
datum of experience” (Einstein, 1934) (also called as ‘Einstein’s razor’, extending the 
Occam’s razor theorem advocating only the simplest solution) warns us to find the 
optimal network representation, which is simple enough, but not too simple. It is an 
important task of the coming years to find the optimal complexity of network 
representation in the drug discovery process. The duality of exploration and 
optimization phases shown on Fig. 19 suggests that network data coverage may be 
extended in consecutive optimization phases including more and more of 
background knowledge.  However, recurrent network simplifications will 
certainly help the ‘surprise factor-aided’ discovery of novel network segments. 
Thomas Singer wrote a few years ago “Extrapolation of preclinical data into 
clinical reality is a translational science and remains an ultimate challenge in drug 
development.” (Signer, 2007). Giving an answer to this challenge our third and last 
major conclusion stresses the importance of network prediction of those human 
data, which are not available experimentally. There are three major components of 
the ‘curse of attrition’ (Fig. 2; Brown & Superti-Furga, 2003; Austin, 2006; Bunnage, 
2011; Ledford, 2012), which are all from this category: 1.) insufficient drug efficacy; 
2.) unexpected major adverse effects; 3.) unexpected forms of human toxicity. All 
these three phenomena are systems-level responses, and the unexpectedness often 
comes from the inability of our logical mind to comprehend the complexity of human 
cells. Network analysis enables a much better design of efficacy taking into account 
patient-, disease stage-, age-specificities (Section 4.3.1.); a better prediction of side-
effects (Section 4.3.5.) and predictive human toxicology (Section 4.3.3.; Henney & 
Superti-Furga, 2008). 
Network science is a novel area of biology; and this is particularly the case with 
respect to drug design. We often lack rigorous comparisons of existing methods, 
which could have allowed a more critical approach to some of them. It is an ongoing 

 
94
effort of the current years to develop benchmarks, gold-standards and rigorous 
assessment tools in network science.  
At the same time, subjectively, we love networks. This approach gave us a 
broader understanding of our complex world in the last decade. We would like to 
share this enthusiasm and our belief that the network approach will greatly help drug 
design of the coming years. 
 
6.2. Systems-level hallmarks of drug quality and trends of network-aided drug 
development helping to achieve them 
In this closing section we identify the systems-level hallmarks of drug quality, 
and list the major trends of network-aided drug development helping to achieve them. 
From the network point of view we need to discriminate between two strategies in 
finding drug targets: 1.) Strategy A aiming to destroy the network of infectious agents 
or cancer cells and 2.) Strategy B aiming to shift the network dynamics of polygenic, 
complex diseases back to normal (Fig. 16; Sections 4.1.1. and 4.1.7.). Both strategies 
converge to the same level of network complexity in hit finding, hit expansion, lead 
selection and optimization phases. 
Table 12 lists the systems-level hallmarks of drug target identification and 
validation, hit finding and development, as well as lead selection and optimization. 
We believe that the systematic application of these systems-level hallmarks will not 
only help the identification of novel drug targets, but will also streamline the drug 
design process to be more selective, less attrition-prone and more profitable.  
We also listed the most important network-related drug design trends helping the 
accomplishment of various systems-level hallmarks. We highlight the development of 
edgetic drugs (Section 4.1.2.), multi-target drugs (Section 4.1.5.) and allo-network 
drugs (Section 4.1.6.) among the richness of network strategies to find novel drug 
targets. We believe that there are a large number of unexplored drug targets, which 
are the hidden masterminds of cellular regulation. Analysis of network dynamics can 
help to find them. Incorporation of disease-stage, age-, gender- and human 
population-specific genetic, metabolome, phosphoproteome and gut microbiome data; 
the development of human ADME and toxicity network models; and the use of side-
effect networks to judge drug safety, may greatly increase the efficiency of the drug 
development process. 
Network-related methods – if applied systematically (and carefully) – will 
uncover a number of novel drug targets, and will increase the efficiency of the drug 
development process. Analysis of the structure and dynamics of molecular networks, 
extended by the network dynamics of constituting proteins and in particular their 
binding sites, provides a novel paradigm of drug discovery. 
 

 
95
Acknowledgments 
 
Authors thank Aditya Barve and Andreas Wagner (University of Zürich, 
Switzerland) for sharing the human homology of enzymes encoding superessential 
metabolic reactions, Haiyuan Yu, Xiujuan Wang (Department of Biological Statistics 
and Computational Biology, Weill Institute for Cell and Molecular Biology, Cornell 
University, Ithaka NY, USA) and Balázs Papp (Szeged Biological Centre, Hungarian 
Academy of Sciences, Szeged, Hungary) for the critical reading of Sections 1.3.3. and 
3.6., respectively. Authors thank Zoltán P. Spiró (École Polytechnique Federale de 
Lausanne, Switzerland) for help in drawing Fig. 11, and members of the LINK-Group 
(
www.linkgroup.hu
) for valuable suggestions. Work in the authors’ laboratory was 
supported by research grants from the Hungarian National Science Foundation 
(OTKA K83314), by the EU (TÁMOP-4.2.2/B-10/1-2010-0013) by a Bolyai 
Fellowship of the Hungarian Academy of Sciences (TK) and by a residence at the 
Rockefeller Foundation Bellagio Center (PC). This project has been funded, in part, 
with federal funds from the NCI, NIH, under contract HHSN261200800001E. This 
research was supported, in part, by the Intramural Research Program of the NIH, 
National Cancer Institute, Center for Cancer Research. The content of this publication 
does not necessarily reflect the views or policies of the Department of Health and 
Human Services, nor does mention of trade names, commercial products, or 
organizations imply endorsement by the U.S. Government. 
 
Conflict of interest statement 
 
The authors declare that there are no conflicts of interest.

 
96
References 
 
Abdi, A., Tahoori, M. B. & Emamian, E. S. (2008). Fault diagnosis engineering of digital circuits can 
identify vulnerable molecules in complex cellular pathways. Science Signaling, 1, ra10. 
Acharyya, S., Oskarsson, T., Vanharanta, S., Malladi, S., Kim, J., Morris, P. G., Manova-Todorova, 
K., Leversha, M., Hogg, N., Seshan, V. E., Norton, L., Brogi, E. & Massagué, J. (2012). A 
CXCL1 paracrine network links cancer chemoresistance and metastasis. Cell, 150, 165-178. 
Adamcsek, B., Palla, G., Farkas, I. J., Derenyi, I. & Vicsek, T. (2006). CFinder: locating cliques and 
overlapping modules in biological networks. Bioinformatics, 22, 1021-1023. 
Adams, J. C., Keiser, M. J., Basuino, L., Chambers, H. F., Lee, D. S., Wiest, O. G. & Babbitt, P. C. 
(2009). A mapping of drug space from the viewpoint of small molecule metabolism. PLoS 
Comput Biol, 5, e1000474. 
Adar, E. (2006). GUESS: A language and interface for graph exploration. In: R. Grinter, T. Rodden, P. 
Aoki, E. Cutrell, R. Jeffries & G. Olson (Eds.), CHI '06 Proceedings of the SIGCHI 
conference on human factors in computing systems. (pp. 791-800). New York: Association 
for Computing Machinery. 
Agoston, V., Csermely, P. & Pongor, S. (2005). Multiple, weak hits confuse complex systems: a 
transcriptional regulatory network as an example. Phys Rev E, 71, 051909. 
Agrawal, S., Dimitrova, N., Nathan, P., Udayakumar, K., Lakshmi, S. S., Sriram, S., Manjusha, N., & 
Sengupta, U. (2008). T2D-Db: an integrated platform to study the molecular basis of type 2 
diabetes. BMC Genomics, 9, 320. 
Agüero, F., Al-Lazikani, B., Aslett, M., Berriman, M., Buckner, F. S., Campbell, R. K., Carmona, S., 
Carruthers, I. M., Chan, A. W., Chen, F., Crowther, G. J., Doyle, M. A., Hertz-Fowler, C., 
Hopkins, A. L., McAllister, G., Nwaka, S., Overington, J. P., Pain, A., Paolini, G. V., Pieper, 
U., Ralph, S. A., Riechers, A., Roos, D. S., Sali, A., Shanmugam, D., Suzuki, T., Van 
Voorhis, W. C. & Verlinde, C. L. (2008). Genomic-scale prioritization of drug targets: the 
TDR Targets database. Nature Rev Drug Discov, 7, 900-907. 
Ahmed, A. & Xing, E. P. (2009). Recovering time-varying networks of dependencies in social and 
biological studies. Proc Natl Acad Sci USA, 106, 11878-11883. 
Ahmed, A., Dwyer, T., Forster, M., Fu, X., Ho, J., Hong, S.-H., Koschäutzki, D., Murray, C., Nikolov, 
N. S., Taib, R., Tarassov, A. & Xu, K. (2006). GEOMI: geometry for maximum insight. 
Lecture Notes Comput Sci, 3843, 468-479. 
Ahn, Y.-Y., Bagrow, J. P. & Lehmann, S. (2010). Link communities reveal multi-scale complexity in 
networks. Nature, 466, 761-764. 
Ajay, A., Walters, W. P. & Murcko, M. A. (1998). Can we learn to distinguish between "drug-like" 
and "nondrug-like" molecules? J Med Chem, 41, 3314-3324. 
Akula, N., Baranova, A., Seto, D., Solka, J., Nalls, M. A., Singleton, A., Ferrucci, L., Tanaka, T., 
Bandinelli, S., Cho, Y. S., Kim, Y. J., Lee, J. Y., Han, B. G., & McMahon, F. J. (2011). A 
network-based approach to prioritize results from genome-wide association studies. PLoS 
ONE, 6, e24220. 
Akutsu, T., Miyano, S. & Kuhara, S. (1999). Identification of genetic networks from a small number of 
gene expression patterns under the Boolean network model. Pac Symp Biocomput, 1999, 17-
28. 
Albert, I. & Albert, R. (2004). Conserved network motifs allow protein-protein interaction prediction. 
Bioinformatics, 20, 3346-3352. 
Albert, R., Jeong, H. & Barabasi, A. L. (2000). Error and attack tolerance of complex networks. 
Nature, 406, 378-382. 
Albert, I., Thakar, J., Li, S., Zhang, R. & Albert, R. (2008). Boolean network simulations for life 
scientists. Source Code Biol Med, 3, 16. 
Alexiou, P., Vergoulis, T., Gleditzsch, M., Prekas, G., Dalamagas, T., Megraw, M., Grosse, I., Sellis, 
T. & Hatzigeorgiou, A. G. (2010). miRGen 2.0: a database of microRNA genomic 
information and regulation. Nucleic Acids Res, 38, D137-D141. 
Ali, M. A. & Sjoblom, T. (2009). Molecular pathways in tumor progression: from discovery to 
functional understanding. Mol Biosyst, 5, 902-908. 
Allard, M., Lebre, V., Robine, J.-M. & Calment, J. (1998). Jeanne Calment: From Van Gogh's time to 
ours: 122 extraordinary years. New York: W.H. Freeman. 
Almaas, E., Kovacs, B., Vicsek, T., Oltvai, Z. N., & Barabasi, A. L. (2004). Global organization of 
metabolic fluxes in the bacterium Escherichia coli. Nature, 427, 839-843. 
Almaas, E., Oltvai, Z. N., & Barabasi, A. L. (2005). The activity reaction core and plasticity of 

 
97
metabolic networks. PLoS Comput Biol, 1, e68. 
Alonso, A., Sasin, J., Bottini, N., Friedberg, I., Osterman, A., Godzik, A., Hunter, T., Dixon, J. & 
Mustelin, T. (2004). Protein tyrosine phosphatases in the human genome. Cell, 117, 699-711. 
Altay, G. (2012). Empirically determining the sample size for large-scale gene network inference 
algorithms. IET Syst Biol, 6, 35-43. 
Alves, N. A. & Martinez, A. S. (2007). Inferring topological features of proteins from amino acid 
residue networks. Physica A, 375, 336-344. 
Andersson, C. D., Chen, B. Y. & Linusson, A. (2010). Mapping of ligand-binding cavities in proteins. 
Proteins, 78, 1408-1422.  
Annibale, A. & Coolen, A. C. C. (2011). What you see is not what you get: how sampling affects 
macroscopic features of biological networks. Interface Focus 1, 836-856. 
Antal, M. A., Böde, C. & Csermely, P. (2009). Perturbation waves in proteins and protein networks: 
Applications of percolation and game theories in signaling and drug design. Curr Prot Pept 
Sci, 10, 161-172. 
Antonov, A. V., Dietmann, S., Rodchenkov, I. & Mewes, H. W. (2009). PPI spider: a tool for the 
interpretation of proteomics data in the context of protein-protein interaction networks. 
Proteomics, 9, 2740-2749. 
Apel, A., Zentgraf, H., Buchler, M. W. & Herr, I. (2009). Autophagy – A double-edged sword in 
oncology. Int J Cancer, 125, 991-995. 
Apic, G., Ignjatovic, T., Boyer, S., & Russell, R. B. (2005). Illuminating drug discovery with 
biological pathways. FEBS Lett, 579, 1872-1877. 
Arkin, M. R., & Wells, J. A. (2004). Small-molecule inhibitors of protein-protein interactions: 
progressing towards the dream. Nat Rev Drug Discov, 3, 301-317. 
Arrell, D. K. & Terzic, A. (2010). Network systems biology for drug discovery. Clin Pharmacol Ther, 
88, 120-125. 
Artymiuk, P. J., Rice, D. W., Mitchell, E. M. & Willett, P. (1990). Structural resemblance between the 
families of bacterial signal-transduction proteins and of G proteins revealed by graph 
theoretical techniques. Protein Eng Des Sel, 4, 39-43. 
Assenov, Y., Ramirez, F., Schelhorn, S. E., Lengauer, T. & Albrecht, M. (2008). Computing 
topological parameters of biological networks. Bioinformatics, 24, 282-284. 
Atias, N., & Sharan, R. (2011). An algorithmic framework for predicting side effects of drugs. J 
Comput Biol, 18, 207-218. 
Atilgan, A. R., Akan, P. & Baysal, C. (2004). Small-world communication of residues and significance 
for protein dynamics. Biophys J, 86, 85-91. 
Audouze, K., Juncker, A. S., Roque, F. J., Krysiak-Baltyn, K., Weinhold, N., Taboureau, O., Jensen, 
T. S., & Brunak, S. (2010). Deciphering diseases and biological targets for environmental 
chemicals using toxicogenomics networks. PLoS Comput Biol, 6, e1000788. 
Austin, O. H. (2006). Research and development in pharmaceutical industry. A Congressional Budget 
Office Study. 
http://www.cbo.gov/publication/18176
.  
Avin C., Lotker, Z. & Pignolet, Y-A. (2011). Structural properties of rich clubs. 
http://arxiv.org/abs/1111.3374
. 
Awan, A., Bari, H., Yan, F., Moksong, S., Yang, S., Chowdhury, S., Cui, Q., Yu, Z., Purisima, E. O., 
& Wang, E. (2007). Regulatory network motifs and hotspots of cancer genes in a mammalian 
cellular signalling network. IET Syst Biol, 1, 292-297. 
Ay, F., Kellis, M. & Kahveci, T. (2011). SubMAP: aligning metabolic pathways with subnetwork 
mappings. J Comput Biol, 18, 219-235. 
Ay, F., Dang, M. & Kahveci, T. (2012). Metabolic network alignment in large scale by network 
compression. BMC Bioinformatics, 13, S2. 
Azmi, A. S., Wang, Z., Philip, P. A., Mohammad, R. M. & Sarkar, F. H. (2010). Proof of concept: 
network and systems biology approaches aid in the discovery of potent anticancer drug 
combinations. Mol Cancer Ther, 9, 3137-3144. 
Azuaje, F., Devaux, Y. & Wagner, D. R. (2010). Identification of potential targets in biological 
signalling systems through network perturbation analysis. BioSystems, 100, 55-64. 
Azuaje, F. J., Zhang, L., Devaux, Y. & Wagner, D. R. (2011). Drug-target network in myocardial 
infarction reveals multiple side effects of unrelated drugs. Sci Rep, 1, 52. 
Bader, G. D., Cary, M. P. & Sander, C. (2006). Pathguide: a pathway resource list. Nucleic Acids Res, 
34, D504-D506. 
Baggs, J. E., Hughes, M. E. & Hogenesch, J. B. (2010). The network as the target. Wiley Interdiscip 
Rev Syst Biol Med, 2, 127-133. 

 
98
Bagler, G. & Sinha, S. (2005). Network properties of protein structures. Physica A, 346, 27-33. 
Baitaluk, M., Kozhenkov, S., Dubinina, Y. & Ponomarenko, J. (2012). IntegromeDB: an integrated 
system and biological search engine. BMC Genomics, 13, 35. 
Bajorath, J., Peltason, L., Wawer, M., Guha, R., Lajiness, M. S. & Van Drie, J. H. (2009). Navigating 
structure-activity landscapes. Drug Discov Today, 14, 698-705. 
Balaji, S., McClendon, C., Chowdhary, R., Liu, J. S. & Zhang, J. (2012). IMID: integrated molecular 
interaction database. Bioinformatics, 28, 747-749. 
Bandyopadhyay, S. & Bhattacharyya, M. (2010). PuTmiR: a database for extracting neighboring 
transcription factors of human microRNAs. BMC Bioinformatics, 11, 190. 
Banerjee, S. J. & Roy, S. (2012). Key to network controllability. 
http://arxiv.org/abs/1209.3737
.  
Barabási, A. L. & Albert, R. (1999). Emergence of scaling in random networks. Science, 286, 509-512. 
Barabási, A. L. & Oltvai, Z. N. (2004). Network biology: understanding the cell's functional 
organization. Nat Rev Genet,